My use of the prediction that I would be able to create a working model of my theories by a certain time enabled me to create a series of predictions of partial achievements which I could use both as benchmarks for the development and as seeds for an ongoing analysis of what went wrong. I think the reasoning of how these predictions enabled me to create these benchmarks should be familiar to anyone who has tried to finish a major undertaking within a certain amount of time.
However, the question of why my latest theory seems to have given me greater hope that I will be able use incremental steps in the development of my AGi project was a little hard to figure out at first. My theory, to refine it a little further, is that the ability to learn effective specializations is a necessary requirement for the development of effective generalizations. But why has this particular theory given me the sense that it may lead to a way to gradually develop my program when my examination of previous efforts to develop AGI seemed to suggest that gradual development was methodologically unsound? There are a number of important aspects to the theory. First, it is a good theory although it might seem a little simplistic. I mean that it makes a lot of sense. Secondly, while people may feel that they have already implicitly incorporated something like the theory into their own theories about AGI, the fact that I highlighted it (in my own mind) is a step that is in some ways similar to formalization. It is a sensible theory and (I feel that) it would be an important part of a formalization of a theory of AGI. For example, a Neural Net enthusiast might claim that Neural Nets were able to incorporate both specializations and generalizations but my criticism of that might be since this process is locked within the complex processes of the Neural Net itself the implicitness of the processes do not make them readily available to the programmer. Because I am more interested in using discernible specializations and generalizations the recognition that these kinds of processes are mutually significant and that one of the challenges in AGI was the achievement of greater generalization, the appreciation of the theory provided me with a new means to break the program down into more fundamental parts. And since I knew that I could write a program that would let me personally define the nature of specializations and of generalizations (in a partially automated program) I realized that I could test different ideas in a simple progression. So when I realized that I could try applying simulations of learned specifications and generalizations I realized that I could test different parts of the theory without having to fully develop the program. So there was something about my appreciation of the nature of thought-derived generalizations that allowed me to develop this new theory. And there was something about the appreciation of the theory that allowed me to break the AGI problem down in a somewhat novel way. And because I realized that I could use these parts as I chose to in an ongoing development of the program I realized that I could use different strategies to develop and test variations in a controlled way. But the development of these ideas will not go smoothly if the theory is not a good one. This analysis gives me some more insight into how problems may be effectively broken into smaller pieces even when previous efforts to do this have run into obstacles. Jim Bromer ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
